Amazon employees are grappling with pressure to use AI tools like MeshClaw, sparking internal debates over productivity metrics and security. The company initially tracked AI adoption through team-wide stats but recently restricted access, allowing only employees and managers to view data. Managers are explicitly barred from tying token usage to performance reviews, per internal guidelines. Meanwhile, tokenmaxxing—the practice of inflating AI engagement to boost rankings—has drawn parallels to similar trends at Meta, where employees game leaderboards using OpenClaw-inspired tools.

MeshClaw, developed by over 30 Amazon staff, automates tasks like code deployments, email triage, and Slack interactions. Internal documents describe it as a “dreaming” agent that consolidates learning overnight and acts on behalf of users during meetings. While Amazon touts the tool as empowering teams to experiment with AI, concerns persist about its security posture. Employees worry the tool’s broad permissions could lead to errors or unintended actions, with one calling the default settings “terrifying.”

The controversy highlights tensions between innovation and accountability. Amazon’s statement emphasizes its commitment to “safe, secure, and responsible” AI development, but critics argue the tool’s design prioritizes efficiency over oversight. The memo detailing MeshClaw’s capabilities underscores its ambition: monitoring deployments and managing workflows autonomously. Yet, as one employee noted, “I’m not about to let it go off and just do its own thing.”

This internal struggle reflects broader industry challenges in balancing AI adoption with ethical guardrails. As tokenmaxxing reshapes workplace dynamics, companies face pressure to align automation tools with human-centric values. For now, Amazon’s approach remains a high-stakes experiment in merging productivity and precaution.

Key takeaway: Amazon’s AI tools, while aimed at streamlining work, reveal the risks of unchecked automation in corporate culture.